After all this mud wrestling lets get back to the original question.
The upcoming Vega chip is supposed to have an updated architecture. Instead of GCN AMD themselves call it now NCU (New Generation Compute Unit). Although they obviously shut down the creativity department there's more behind it than just a few letters. The pipeline is restocked and ready for higher frequencies. The load balancer has improved. They can schedule more work around the CUs, where before small operations could block the whole chip. The engines now can natively process 8-/16-/32- and 64-bit ops in each clock cycle.
The engines are now a client of the onboard L2 cache. By making the render back-end clients of the L2 cache they get access to large pools in a much bigger buffer, the direct effect this has is in the end to improve in performance with applications that use heavy read-after-write operations. So here we should see an efficiency and thus performance increase. And let's not forget HBM2, which is also an improvement over NVidia.
There's more in there, but it should be already clear by now that they did not just a copy/paste job of GCN. Let's see what comes up. I personally like competition and have to agree with epixoip that AMD is not much of an alternative since NVidia came out with Maxwell two and a half years ago. Vega is the first promising chip for AMD in 14nm and could change something, at least to be at par.
And btw. the AMD shares roared up by 350% last year. The signs are there.
The upcoming Vega chip is supposed to have an updated architecture. Instead of GCN AMD themselves call it now NCU (New Generation Compute Unit). Although they obviously shut down the creativity department there's more behind it than just a few letters. The pipeline is restocked and ready for higher frequencies. The load balancer has improved. They can schedule more work around the CUs, where before small operations could block the whole chip. The engines now can natively process 8-/16-/32- and 64-bit ops in each clock cycle.
The engines are now a client of the onboard L2 cache. By making the render back-end clients of the L2 cache they get access to large pools in a much bigger buffer, the direct effect this has is in the end to improve in performance with applications that use heavy read-after-write operations. So here we should see an efficiency and thus performance increase. And let's not forget HBM2, which is also an improvement over NVidia.
There's more in there, but it should be already clear by now that they did not just a copy/paste job of GCN. Let's see what comes up. I personally like competition and have to agree with epixoip that AMD is not much of an alternative since NVidia came out with Maxwell two and a half years ago. Vega is the first promising chip for AMD in 14nm and could change something, at least to be at par.
And btw. the AMD shares roared up by 350% last year. The signs are there.